Over-thinking NAS configuration, again

[porina]

Currently:

NAS1: 9x 8TB Drives, Unraid 2 parity + 7 data - 56TB usable, two disk failure recoverable. 36TB in use.

NAS2: 1x 6TB, 4x 3TB, Unraid no parity, all data - 18TB usable, no redundancy.

 

NAS2 only exists to make a backup of the data on NAS1. That is why it is run without redundancy. Unraid has the feature that you only lose the data on a disk if it fails, but surviving disks will still have data intact. In the event a disk fails on NAS2, I will lose that backup copy of whatever was on it, but I can re-run the backup from NAS1 and it will recreate it on remaining disks, providing there is sufficient capacity.

 

You might spot one major flaw in this current arrangement. NAS2 is less than half the size of NAS1 currently! I'm not using all the space on NAS1, 36TB of data on it is about double that of NAS2 capacity. For now, I'm doing a selective backup of most important data first. This was only recently started and it could take many days to sync the data.

 

I'm wondering, do I really need two disk redundancy on NAS1? I know, raid is not backup. While NAS1 was my only copy of that data, it gave me more confidence against mechanical disk failure. But If I have a backup copy elsewhere, then it becomes less critical. I'm thinking I could go down to 1 disk redundancy and that freed disk could go into NAS2 for more backup capacity. If I go down to one parity drive in NAS1, then to lose data I'd have to have 2 or more simultaneous disk failures and the data loss would only happen if the data on the lost disk(s) was not yet copied onto NAS2.

 

I forgot to say, I'm trying not to spend more here! I could in theory move more currently unused data capacity from NAS1 to NAS2, but I'm hesitant to do this as they don't provide an easy and clean solution to remove data disks, only to replace or add them.

 

I've noticed lower drive capacity pricing has been steadily going up. Makes me hesitant to get more around 8TB level. Best capacity value seems to be around the 16TB area. One of these would give NAS2 just enough capacity to backup NAS1.

 

Another side thought. NAS2 has each disk running independent ZFS. Not RaidZ, but one logical ZFS system per disk. Unraid calls this hybrid mode. ZFS has compression support which I have enabled. Last time I checked stats it was averaging about 1% saving. Not great, but not nothing, however close it is. I'd say the bulk of the data is photos, images (in compressed formats), and video, so I guess this isn't really surprising.

 

Edit: compression working better than expected!

[OddOod]

ooof, that's a LOT of little drives. I'd recommend stacking some cash and grabbing some refurbed 16s off serverpartdeals 

[porina]

5 minutes ago, OddOod said:

ooof, that's a LOT of little drives. I'd recommend stacking some cash and grabbing some refurbed 16s off serverpartdeals 

I'm in the UK. Cheap used/refurb drives don't exist. The ones that do exist are at best 2/3 the cost of new, but 90 day warranty and unknown usage. All the cheap "new" drives I suspect are used Seagate drives with faked SMART and label since they're all from no name shops.

 

I almost ditched the 3TB drives but they're still fine for now. At least with Unraid, when the time does come to swap them with higher capacity, it'll be easy.

[OddOod]

26 minutes ago, porina said:

The ones that do exist are at best 2/3 the cost of new

TBH, that's about what I pay in the US for manufacturer recerts from SPD. I have 32 14TB drives split between a pair of NASs (in 8 disk Z1 stripes so I can get ~200TB of nominal storage) and they have been performing flawlessly for 4 years now. 
I pulled SMART on each of them and all but 2 had ~20 hours of spin time. My guess is that those are just drives had control board failures. The other two were at 10k hours (~1 year) and were likely incorrectly returned, probably threw errors due to bad sata connection. 
The huge advantage of buying recerts is that they tend to be from different batches, so mass failures are super unlikely.

[porina]

3 hours ago, OddOod said:

TBH, that's about what I pay in the US for manufacturer recerts from SPD.

These were seller certified. I'm going to guess they were pulls so high usage. They never responded when I asked them in the past.

 

Had another look, I can't even find the ones I saw before now. The used market is pretty dry. I saw a theory that sounds plausible, that due to stricter requirements on data privacy, big companies here would rather destroy than wipe and resell disks.

 

There's also a claim that AI is gobbling up disk supply so other areas may be in constraint with subsequent price rises. Might have to bite the bullet and get some more new ones before they go up more. For indication, most of my 8TB ones are SMR bought in last couple of years around £135 each. At same capacity, I missed out on Enterprise ones around £145 a couple months ago, and they've never returned in stock at any price. Even the more expensive enterprise/NAS options around £160 are gone now. Best I can do today are around £190.

 

Larger capacity ones don't seem too common in consumer supply chain although the per capacity cost of 16TB tends to be a bit better than at 8TB. It would mess with my existing arrays though and I'm not burning a ton of cash to replace them wholesale. I waited and it will cost me with no expected improvement in near future.

[OddOod]

1 hour ago, porina said:

big companies here would rather destroy than wipe and resell disks.

Oh, absolutely. Even in the US they'd rather be safe than sorry. Which is why I don't see many enterprise castoffs on the market. It's all private users or manufacturer rexerts. 
Entertainingly, even the DOD agrees that a single random write pass is sufficient to securely erase a disk, though not sure companies want to spend the time securely wiping drives, even if they can do it at 1TB/hr

[porina]

9 minutes ago, OddOod said:

Entertainingly, even the DOD agrees that a single random write pass is sufficient to securely erase a disk, though not sure companies want to spend the time securely wiping drives, even if they can do it at 1TB/hr

1TB/h is about 278 MB/s. Can higher capacity drives sustain that? Even at 1TB/h, a large capacity disk could take around a day each, plus you need to be confident there is no human factor leading to unwiped disks slipping out. Would the resale value offset implementing and running that operation? Mechanical destruction is fast and you know it is done.

 

I had wondered about recovery after overwrite. When I sell HDs, I do a single overwrite pass and consider it done. With SSDs, I do a full TRIM although I don't verify. I wish SSDs were deterministic TRIM as standard operation as that also would be good enough for me, but maybe not for the intelligence services. The only reason I don't do a secure erase is I never got it working! Good for preventing accidental triggering, bad for actually wanting to use it.

 

In the distant past it was claimed it was due to slight positioning shifts meaning tracks weren't fully overwritten, and with a sufficiently sensitive read head you could selectively read the non-overwritten part. Another similar claim was that due to the analog nature, they aren't perfect logical 1s and 0s. Writing a 1 on top of 1, was different value than 1 on top of 0, and again that difference was claimed to be detectable. However today's disks have pushed density higher than ever, both in track spacing and also linear density. With so much precision and imperfection, it seems less likely those claimed older techniques could work well if they ever did at all.

 

I guess they also don't trust "fast erase" methods, like use the disk encrypted and throw away the key when done. There'll be that non-zero theoretical chance someone finds a way to recover the key, however unlikely it is if implemented correctly.

[OddOod]

On 10/30/2025 at 2:41 PM, porina said:

1TB/h is about 278 MB/s. Can higher capacity drives sustain that?

Yup. Pulled the data sheet for a SG exos X16 16TB and it's sustained write speed is 250MiB/s

On 10/30/2025 at 2:41 PM, porina said:

no human factor leading to unwiped disks slipping out

Hard agree

On 10/30/2025 at 2:41 PM, porina said:

Would the resale value offset implementing and running that operation?

Deeply unlikely

 

On 10/30/2025 at 2:41 PM, porina said:

When I sell HDs, I do a single overwrite pass and consider it done

TBH, I don't even overwrite. Nothing on those disks has any info I care about. It's all ISOs. And I'm not sure you could even retrieve that if you only have half the RAID array (8 disk Z1)

On 10/30/2025 at 2:41 PM, porina said:

With so much precision and imperfection, it seems less likely those claimed older techniques could work well if they ever did at all.

IIRC they were demonstrated in controlled environments, but they weren't cheap and weren't reliable. But yeah, the march of time has driven increased density which rendered them useless
 

[porina]

44 minutes ago, OddOod said:

Yup. Pulled the data sheet for a SG exos X16 16TB and it's sustained write speed is 250MiB/s

Maximum sustained, not continuous sustained. On outer disk, where it is fastest. It'll drop off, as much as half by the inner part. Whole disk average will be lower.

 

44 minutes ago, OddOod said:

TBH, I don't even overwrite. Nothing on those disks has any info I care about. It's all ISOs. And I'm not sure you could even retrieve that if you only have half the RAID array (8 disk Z1)

Yeah, if you know the exact circumstances it may not be necessary. For example, in the distant past when I still used HDs for games, I wouldn't bother wiping those if all they have were game install files. Basically unless I know a disk either has never had any potentially personal info on it, or has been fully overwritten at some point since it has, then I'd just do a single pass for confidence.

 

 

Ooh, forgot this was my thread. An update then!

 

Since starting the thread, I have done some manual de-duplication, and deletion of anything I no longer feel worth keeping. That reduced my current "to keep" data set to 31.1TB. That allowed me to pull one unused data disk out of NAS1, as well as retiring one of the parity drives since the redundancy is now a 2nd copy on another NAS. I added them to NAS2 and ran into a problem. I'm out of SATA ports and can't add any more without spending somewhere. So I had to retire one of the 3TB drives in order to add the two 8TB ones, for a net gain of 13TB capacity. The indicated capacity of the backup NAS is now 30.2TB. Which is about 1TB short of the used capacity on NAS1. Not quite enough to make a full copy, or is it? NAS2 uses ZFS and I have compression enabled. Depending on the drive and file mix, I'm seeing up to 5% compression. That might be enough to get me that extra capacity. Also, there remains potential for me to de-duplicate more, or delete other things. Just a matter of being motivated enough to spend time doing it.

 

Current status:

NAS1: 7x 8TB Drives, Unraid 1 parity + 6 data - 48TB usable, one disk failure recoverable. 31TB in use.

NAS2: 2x 8TB, 1x 6TB, 3x 3TB, Unraid no parity, all data - 30TB usable, no redundancy.

[OddOod]

1 minute ago, porina said:

Maximum sustained, not continuous sustained. On outer disk, where it is fastest. It'll drop off, as much as half by the inner part. Whole disk average will be lower.

Huh, neat! TIL